test_feature_extraction: use baseline fixture

Signed-off-by: U. Artie Eoff <[email protected]>

tests/baselines/fixture/tests/test_feature_extraction.json

@@ -0,0 +1,10 @@
+{
+  "tests/test_feature_extraction.py::GaudiFeatureExtractionTester::test_latency_graph_bf16": {
+    "gaudi1": {
+      "time_per_iter": 0.7987
+    },
+    "gaudi2": {
+      "time_per_iter": 0.6812
+    }
+  }
+}

tests/test_feature_extraction.py

@@ -94,6 +94,13 @@ class GaudiFeatureExtractionTester(TestCase):
     Tests for Supabase/gte-small feature extraction on Gaudi
     """
 
+    @pytest.fixture(autouse=True)
+    def _use_(self, baseline):
+        """
+        https://docs.pytest.org/en/stable/how-to/unittest.html#using-autouse-fixtures-and-accessing-other-fixtures
+        """
+        self.baseline = baseline
+
     def test_inference_default(self):
         """
         Tests for equivalent CPU and HPU outputs
@@ -135,4 +142,8 @@ def test_latency_graph_bf16(self):
         torch.hpu.synchronize()
         end_time = time.time()
         time_per_iter = (end_time - start_time) * 1000 / test_iters  # time in ms
-        self.assertLess(time_per_iter, 1.05 * LATENCY_GTE_SMALL_BF16_GRAPH_BASELINE)
+        self.baseline.assertRef(
+            compare=lambda actual, ref: actual < (1.05 * ref),
+            context=[OH_TEST_DEVICE],
+            time_per_iter=time_per_iter,
+        )